Research Area - Hyperdimensional Computing (HDC)

ANII Research grant FCE_1_2023_1_176242 (“Nuevas direcciones para el Cómputo Hiperdimensional”).

Feel free to contact us for more information

For decades, artificial neural networks (ANNs) have dominated the landscape of artificial intelligence, mimicking — albeit loosely — the connections between neurons in the brain. But what if there is a different paradigm, one that forgoes the intricate architecture of layered nodes and instead encodes knowledge into abstract symbols? This is the promise of hyperdimensional computing (HDC), a framework in which information is represented as vectors with thousands of dimensions -hypervectors- where information is distributed across dimensions.

Hypervectors are remarkably robust: they maintain their semantic integrity even when many of their components are altered. This resilience arises because meaning is distributed across dimensions, not localized in any single part. This makes HDC inherently fault-tolerant and ideal for low-power, noise-resilient computing — systems where memory and computation can merge without sacrificing stability.

Crucially, this approach reflects how our brains store information: not in single neurons, but in overlapping patterns across many. Similarly, a hypervector spreads meaning across all dimensions, allowing concepts to be combined and manipulated through simple algebra. This means we can define an algebra of operations over hypervectors — a mathematical framework that enables the composition, transformation, and extraction of structured information using well-defined operators.

HDC is a new framework for artificial intelligence that not only covers classical machine learning tasks but also naturally extends into the domain of symbolic AI.

Current research interests in our research group:

  • Developing hypervector-based representations for graphs and networks - Using HDC to model pharmaceutical and repetitive molecular structures.
  • Designing algebraic operators for symbolic reasoning over hypervectors.
  • Representing structured knowledge through distributed hypervectors
  • Investigating attention mechanisms within HDC frameworks.
  • Building interpretable models using hyperdimensional representations.
  • Creating multimodal representations by integrating structured and non-structured data.